Unit 7, lesson 39

Regular Singular Points and Frobenius

How to recognize a regular singular point, build a Frobenius series from the indicial equation, and decide when roots differing by an integer or a repeated indicial root lead to a logarithmic second solution.

  • Video 39
  • Power-Series and Frobenius Solutions

Near x=0x = 0 the equation x2y′′+xy′−49y=0x^2y'' + xy' - \tfrac49 y = 0, with independent variable xx and y′=dy/dxy' = dy/dx, has the solutions x2/3x^{2/3} and x−2/3x^{-2/3}, and the similar equation x2y′′+xy′=0x^2y'' + xy' = 0 has the solution ln⁡x\ln x. No power series in whole powers of xx behaves like these near 00, so the ordinary-point method cannot find them. Which equations still allow a series solution at a singular point, and what form must it take? This lesson classifies regular singular points, finds the leading power from the indicial equation, and shows when the second solution needs a logarithm. Throughout, the work is on the interval x>0x > 0, where xr=erln⁡xx^r = e^{r \ln x} and ln⁡x\ln x are defined without any choice of branch.

Key terms

Regular singular point

Definition Let x0x_0 be a singular point of the normalized equation

y′′+p(x) y′+q(x) y=0,y'' + p(x)\,y' + q(x)\,y = 0,

so that pp or qq is not analytic at x0x_0. The point x0x_0 is a regular singular point if both (x−x0) p(x)(x - x_0)\,p(x) and (x−x0)2 q(x)(x - x_0)^2\,q(x) are analytic at x0x_0, once their values at x0x_0 are defined by their limits. Otherwise x0x_0 is an irregular singular point. For a regular singular point at x0=0x_0 = 0, multiplying the normalized equation by x2x^2 gives the standard form

x2y′′+x P(x) y′+Q(x) y=0,\begin{aligned} &x^2y'' + x\,P(x)\,y' \\[0.4em] &\qquad + Q(x)\,y = 0, \end{aligned}

where P(x)=x p(x)P(x) = x\,p(x) and Q(x)=x2q(x)Q(x) = x^2q(x) are analytic at 00. A regular singular point at another x0x_0 moves to 00 under the substitution s=x−x0s = x - x_0.

In plain words At a regular singular point, pp has at worst a simple pole and qq at worst a double pole. That is the same strength of singularity as in a Cauchy-Euler equation, and it is mild enough for a modified series.

Example For x2y′′+xy′−49y=0x^2y'' + xy' - \tfrac49 y = 0 the normalized coefficients p(x)=1/xp(x) = 1/x and q(x)=−4/(9x2)q(x) = -4/(9x^2) are not analytic at 00, but x p(x)=1x\,p(x) = 1 and x2q(x)=−49x^2q(x) = -\tfrac49 are constants. So 00 is a regular singular point with P=1P = 1 and Q=−49Q = -\tfrac49. Bessel’s equation x2y′′+xy′+(x2−ν2) y=0x^2y'' + xy' + (x^2 - \nu^2)\,y = 0 is regular singular at 00 as well, with P=1P = 1 and Q(x)=x2−ν2Q(x) = x^2 - \nu^2. By contrast, x3y′′+y=0x^3y'' + y = 0 has q(x)=1/x3q(x) = 1/x^3, and x2q(x)=1/xx^2q(x) = 1/x is not analytic at 00, so 00 is an irregular singular point.

Frobenius series

Definition At a regular singular point 00, a Frobenius series is a trial solution on x>0x > 0 of the form

y=xr∑n=0∞anxn=∑n=0∞anxn+r,\begin{aligned} y &= x^r \sum_{n=0}^{\infty} a_n x^n \\[0.4em] &= \sum_{n=0}^{\infty} a_n x^{n+r}, \end{aligned}

with a0≠0a_0 \ne 0, where the exponent rr is to be found and may be fractional or negative. The condition a0≠0a_0 \ne 0 makes xrx^r the actual leading power. The theorem of Frobenius says when such a solution exists. Suppose the series of PP and QQ about 00 converge for ∣x∣<ρ\lvert x \rvert < \rho, and the roots r1≥r2r_1 \ge r_2 of the indicial equation below are real. Then there is a solution y1=xr1∑anxny_1 = x^{r_1}\sum a_n x^n with a0=1a_0 = 1 whose series converges at least for ∣x∣<ρ\lvert x \rvert < \rho, so y1y_1 solves the equation on 0<x<ρ0 < x < \rho. If r1−r2r_1 - r_2 is not an integer, there is a second solution of the same form with exponent r2r_2. On x<0x < 0 the same statements hold with xrx^r replaced by ∣x∣r\lvert x \rvert^r.

In plain words Let the equation choose a leading power xrx^r first, then correct it with an ordinary power series. Close to 00 such a solution behaves like a constant times xrx^r.

Example For x2y′′+xy′−49y=0x^2y'' + xy' - \tfrac49 y = 0 the coefficients PP and QQ are constant, and the Frobenius series reduce to single powers. For any power,

x2(xr)′′+x(xr)′=r2xr,x^2(x^r)'' + x(x^r)' = r^2x^r,

so y=xry = x^r is a solution exactly when r2=49r^2 = \tfrac49, which gives x2/3x^{2/3} and x−2/3x^{-2/3}. The singular point does not accept arbitrary initial values, so data are imposed at an ordinary point such as x=1x = 1. For y=Ax2/3+Bx−2/3y = Ax^{2/3} + Bx^{-2/3} the conditions y(1)=3y(1) = 3 and y′(1)=23y'(1) = \tfrac23 read A+B=3A + B = 3 and 23(A−B)=23\tfrac23(A - B) = \tfrac23, so A=2A = 2, B=1B = 1 and

y=2x2/3+x−2/3.y = 2x^{2/3} + x^{-2/3}.

Indicial equation

Definition Substitute a Frobenius series into x2y′′+x P(x) y′+Q(x) y=0x^2y'' + x\,P(x)\,y' + Q(x)\,y = 0, where P(x)=∑PkxkP(x) = \sum P_k x^k and Q(x)=∑QkxkQ(x) = \sum Q_k x^k. The lowest power xrx^r has the coefficient a0F(r)a_0F(r), with

F(r)=r(r−1)+P(0) r+Q(0).\begin{aligned} F(r) &= r(r - 1) \\[0.4em] &\quad + P(0)\,r + Q(0). \end{aligned}

Since a0≠0a_0 \ne 0, the exponent must satisfy the indicial equation F(r)=0F(r) = 0, and its roots are called the indicial roots, or exponents, at the singular point. For n≥1n \ge 1 the coefficient of xn+rx^{n+r} gives

F(r+n)an=−∑k=0n−1cn,kak,cn,k=(k+r)Pn−k+Qn−k.\begin{gathered} F(r + n)a_n = -\sum_{k=0}^{n-1} c_{n,k}a_k, \\[0.4em] c_{n,k} = (k + r)P_{n-k} + Q_{n-k}. \end{gathered}

This determines ana_n from the earlier coefficients whenever F(r+n)≠0F(r + n) \ne 0. At an index where F(r+n)=0F(r + n) = 0 the equation for ana_n cannot be solved in the usual way, and that is where the difficulties with the second solution come from.

In plain words Very close to 00 only the leading terms matter, and there the equation acts like the Cauchy-Euler equation x2y′′+P(0) xy′+Q(0) y=0x^2y'' + P(0)\,xy' + Q(0)\,y = 0. Its exponents are the possible leading powers of the true solutions.

Example Multiplying xy′′+y=0xy'' + y = 0 by xx gives x2y′′+x y=0x^2y'' + x\,y = 0, so P=0P = 0 and Q(x)=xQ(x) = x, and 00 is a regular singular point. The indicial equation is r(r−1)=0r(r - 1) = 0, with roots 11 and 00. For r=1r = 1 the only nonzero term on the right is the one with Q1=1Q_1 = 1, and F(n+1)=n(n+1)F(n + 1) = n(n + 1), so the recurrence is an=−an−1/(n(n+1))a_n = -a_{n-1}/(n(n + 1)). With a0=1a_0 = 1 it gives

y1=∑n=0∞(−1)nxn+1n! (n+1)!=x−x22+x312−⋯ ,\begin{aligned} y_1 &= \sum_{n=0}^{\infty} \frac{(-1)^n x^{n+1}}{n!\,(n+1)!} \\[0.4em] &= x - \frac{x^2}{2} + \frac{x^3}{12} - \cdots, \end{aligned}

which converges for every xx, since PP and QQ are polynomials.

Roots differing by an integer

Definition Let the indicial roots be real with r1−r2=Nr_1 - r_2 = N for a positive integer NN. The larger root always gives a Frobenius solution y1y_1, because F(r1+n)≠0F(r_1 + n) \ne 0 for every n≥1n \ge 1. For the smaller root the recurrence reaches n=Nn = N with F(r2+N)=F(r1)=0F(r_2 + N) = F(r_1) = 0, so the equation for aNa_N reads 0⋅aN=SN0 \cdot a_N = S_N, where SNS_N is the right side computed from a0,…,aN−1a_0, \dots, a_{N-1}. If SN≠0S_N \ne 0, no Frobenius solution with exponent r2r_2 exists, and the second solution contains a logarithm. If SN=0S_N = 0, every value of aNa_N works, and r2r_2 gives a second Frobenius solution with no logarithm. Changing aNa_N only adds a multiple of y1y_1. When r1−r2r_1 - r_2 is not an integer, both roots give Frobenius solutions, and they are independent because their leading powers differ. An integer difference is therefore a reason to check the equation at n=Nn = N, not a proof that a logarithm appears.

In plain words When the two leading powers are a whole number apart, the series that starts at the lower power eventually reaches the higher one. At that step the equation either lets it pass or forces a logarithm.

Example For x2y′′+xy′−14y=0x^2y'' + xy' - \tfrac14 y = 0 the roots are 12\tfrac12 and −12-\tfrac12, so N=1N = 1, yet x1/2x^{1/2} and x−1/2x^{-1/2} both solve it. Here PP and QQ are constant, so S1=0S_1 = 0. For xy′′+y=0xy'' + y = 0 the roots 11 and 00 also differ by 11, but with r2=0r_2 = 0 the equation at n=1n = 1 is F(1) a1=−Q1a0F(1)\,a_1 = -Q_1a_0, that is 0⋅a1=−a00 \cdot a_1 = -a_0. This is impossible because a0≠0a_0 \ne 0, so that equation has no second Frobenius solution, and its second solution must contain ln⁡x\ln x.

Repeated indicial root

Definition The indicial root r1r_1 is repeated when F(r)=(r−r1)2F(r) = (r - r_1)^2. Then F(r1+n)=n2≠0F(r_1 + n) = n^2 \ne 0 for every n≥1n \ge 1, so the recurrence determines every coefficient from a0a_0, and every Frobenius solution with exponent r1r_1 is a constant multiple of y1y_1. A second independent solution therefore cannot be a Frobenius series. It has the form

y2=y1ln⁡x+xr1∑n=1∞bnxny_2 = y_1 \ln x + x^{r_1} \sum_{n=1}^{\infty} b_n x^n

on 0<x<ρ0 < x < \rho, so in this case the logarithm is always present.

In plain words A double root supplies only one leading power, so the second solution has to come from somewhere else. It comes from a factor ln⁡x\ln x, just as a double root of a Cauchy-Euler equation gives xr1x^{r_1} and xr1ln⁡xx^{r_1}\ln x.

Example The equation x2y′′+xy′=0x^2y'' + xy' = 0 has P=1P = 1 and Q=0Q = 0, so F(r)=r2F(r) = r^2 and 00 is a double root. The constant y1=1y_1 = 1 is a solution. With v=y′v = y' the equation becomes xv′+v=0xv' + v = 0, so v=C/xv = C/x and a second solution is y2=ln⁡xy_2 = \ln x. Its derivatives 1/x1/x and −1/x2-1/x^2 give x2⋅(−1/x2)+x⋅(1/x)=0x^2 \cdot (-1/x^2) + x \cdot (1/x) = 0. The Wronskian W(1,ln⁡x)=1⋅(1/x)−0⋅ln⁡x=1/xW(1, \ln x) = 1 \cdot (1/x) - 0 \cdot \ln x = 1/x is nonzero on x>0x > 0, so 11 and ln⁡x\ln x form a fundamental set there. In the form above, every bnb_n is zero.

Logarithmic second solution

Definition Let 00 be a regular singular point with real indicial roots r1≥r2r_1 \ge r_2, and let the series of PP and QQ converge for ∣x∣<ρ\lvert x \rvert < \rho. If the roots are equal, a second solution on 0<x<ρ0 < x < \rho has the form shown for a repeated indicial root, with the coefficient of y1ln⁡xy_1 \ln x equal to 11. If r1−r2r_1 - r_2 is a positive integer, a second solution on 0<x<ρ0 < x < \rho is

y2=C y1ln⁡x+xr2∑n=0∞bnxn\begin{aligned} y_2 &= C\,y_1 \ln x \\[0.4em] &\quad + x^{r_2} \sum_{n=0}^{\infty} b_n x^n \end{aligned}

with b0≠0b_0 \ne 0, where the constant CC is zero exactly when the smaller root gives a Frobenius solution. In both cases the series converge at least for ∣x∣<ρ\lvert x \rvert < \rho, and y1y_1 and y2y_2 form a fundamental set on 0<x<ρ0 < x < \rho. The coefficients are found by substituting the form into the equation. Alternatively, on an interval where y1y_1 has no zeros, reduction of order gives

y2=y1∫e−∫p dxy12 dx,y_2 = y_1 \int \frac{e^{-\int p\,dx}}{y_1^2}\,dx,

and the logarithm appears when the integrand contains a term 1/x1/x. On x<0x < 0 the same forms hold with ∣x∣\lvert x \rvert in place of xx.

In plain words When the second leading power is missing or blocked, the equation makes up the shortfall with ln⁡x\ln x times the first solution, plus a correcting series.

Example For x2y′′−xy′+y=0x^2y'' - xy' + y = 0 the indicial polynomial is r(r−1)−r+1=(r−1)2r(r - 1) - r + 1 = (r - 1)^2, so r=1r = 1 is repeated and y1=xy_1 = x. Here p(x)=−1/xp(x) = -1/x and e−∫p dx=xe^{-\int p\,dx} = x, so reduction of order gives

y2=x∫xx2 dx=xln⁡x,y_2 = x \int \frac{x}{x^2}\,dx = x \ln x,

with the logarithm coming from ∫dx/x\int dx/x. For xy′′+y=0xy'' + y = 0, where the smaller root 00 is blocked, substituting the second form with b0=1b_0 = 1 and the free choice b1=0b_1 = 0 gives

y2=−y1ln⁡x+1−34x2+736x3−⋯ ,\begin{aligned} y_2 &= -y_1 \ln x + 1 - \tfrac{3}{4}x^2 \\[0.4em] &\quad + \tfrac{7}{36}x^3 - \cdots, \end{aligned}

with y1y_1 from the indicial equation card. The constant is forced, since the constant term of xy2′′+y2xy_2'' + y_2 works out to C+b0C + b_0, so C=−1C = -1.

Check yourself

  1. Classify the point x=0x = 0 for x2y′′+3xy′+(1+x) y=0x^2y'' + 3xy' + (1 + x)\,y = 0 and for x3y′′+xy′+y=0x^3y'' + xy' + y = 0. For the regular singular one, find the indicial roots and say what they predict about the second solution.
Show the answer

For the first equation p(x)=3/xp(x) = 3/x and q(x)=(1+x)/x2q(x) = (1 + x)/x^2, so 00 is singular, but x p(x)=3x\,p(x) = 3 and x2q(x)=1+xx^2q(x) = 1 + x are analytic, so 00 is a regular singular point with P=3P = 3 and Q(x)=1+xQ(x) = 1 + x. The indicial polynomial is

r(r−1)+3r+1=(r+1)2,\begin{aligned} &r(r - 1) + 3r + 1 \\[0.4em] &\qquad = (r + 1)^2, \end{aligned}

so −1-1 is a repeated root. Up to a constant factor there is one Frobenius solution with exponent −1-1, which works out to y1=x−1∑(−1)nxn/(n!)2y_1 = x^{-1}\sum (-1)^n x^n/(n!)^2, and the second solution must contain y1ln⁡xy_1 \ln x. For the second equation p(x)=1/x2p(x) = 1/x^2, and x p(x)=1/xx\,p(x) = 1/x is not analytic at 00, so 00 is an irregular singular point and the Frobenius theory does not apply there.

  1. Find the indicial roots of 2x2y′′+3xy′−(1+x) y=02x^2y'' + 3xy' - (1 + x)\,y = 0 at 00. How many Frobenius solutions are guaranteed? Find the first three coefficients of each with a0=1a_0 = 1.
Show the answer

Dividing by 22 gives the standard form with P=32P = \tfrac32 and Q(x)=−12(1+x)Q(x) = -\tfrac12(1 + x), so

F(r)=r(r−1)+32r−12=(r+1)(r−12).\begin{aligned} F(r) &= r(r - 1) + \tfrac32 r - \tfrac12 \\[0.4em] &= (r + 1)(r - \tfrac12). \end{aligned}

The roots 12\tfrac12 and −1-1 differ by 32\tfrac32, which is not an integer, so there are two independent Frobenius solutions, and since PP and QQ are polynomials their series converge for all xx. In the original equation the coefficient of xn+rx^{n+r} is (2(n+r)−1)(n+r+1) an−an−1(2(n + r) - 1)(n + r + 1)\,a_n - a_{n-1}. For r=12r = \tfrac12 this gives an=an−1/(n(2n+3))a_n = a_{n-1}/(n(2n + 3)), so a1=15a_1 = \tfrac15, a2=170a_2 = \tfrac{1}{70} and

y1=x1/2(1+x5+x270+⋯ ).y_1 = x^{1/2}(1 + \tfrac{x}{5} + \tfrac{x^2}{70} + \cdots).

For r=−1r = -1 it gives an=an−1/(n(2n−3))a_n = a_{n-1}/(n(2n - 3)), so a1=−1a_1 = -1, a2=−12a_2 = -\tfrac12 and

y2=x−1(1−x−x22−⋯ ).y_2 = x^{-1}(1 - x - \tfrac{x^2}{2} - \cdots).
  1. The indicial roots of xy′′+2y′+xy=0xy'' + 2y' + xy = 0 at 00 differ by an integer, and a classmate concludes that the second solution must contain ln⁡x\ln x. Check this against the recurrence and identify both solutions.
Show the answer

Multiplying by xx gives x2y′′+2xy′+x2y=0x^2y'' + 2xy' + x^2y = 0, so P=2P = 2 and Q(x)=x2Q(x) = x^2. The indicial equation r(r−1)+2r=r(r+1)=0r(r - 1) + 2r = r(r + 1) = 0 has the roots 00 and −1-1, with N=1N = 1. The coefficient of xn+rx^{n+r} is F(n+r) an+an−2F(n + r)\,a_n + a_{n-2}, with a−2=a−1=0a_{-2} = a_{-1} = 0. For r=−1r = -1 and n=1n = 1 this reads F(0) a1=0F(0)\,a_1 = 0, that is 0⋅a1=00 \cdot a_1 = 0, so there is no obstruction and a1a_1 is free. The classmate is wrong. With a1=0a_1 = 0 the recurrence an=−an−2/(n(n−1))a_n = -a_{n-2}/(n(n - 1)) makes every odd coefficient zero and gives a2=−12a_2 = -\tfrac12, a4=124a_4 = \tfrac{1}{24} and in general a2k=(−1)k/(2k)!a_{2k} = (-1)^k/(2k)!, so

y=∑k=0∞(−1)kx2k−1(2k)!=cos⁡xx.\begin{aligned} y &= \sum_{k=0}^{\infty} \frac{(-1)^k x^{2k-1}}{(2k)!} \\[0.4em] &= \frac{\cos x}{x}. \end{aligned}

The root 00, with an=−an−2/(n(n+1))a_n = -a_{n-2}/(n(n + 1)) and a1=0a_1 = 0, gives a2k=(−1)k/(2k+1)!a_{2k} = (-1)^k/(2k + 1)! and the solution sin⁡x/x\sin x / x. A direct check confirms both, because the equation can be written (xy)′′+xy=0(xy)'' + xy = 0, so xyxy is a combination of cos⁡x\cos x and sin⁡x\sin x. Neither solution has a logarithm.

How this lesson connects